Counting trees in Random Forests: Predicting symptom severity in psychiatric intake reports.

Area of Science:

• Clinical Natural Language Processing
• Medical Informatics
• Psychiatric Diagnosis

Background:

• Clinical Natural Language Processing (NLP) tasks often require extracting specific information from unstructured clinical notes.
• Assigning severity scores to psychiatric symptoms is crucial for patient assessment and treatment planning.
• Existing methods may not fully leverage the inherent structure of psychiatric intake reports.

Purpose of the Study:

• To develop and evaluate a novel method for assigning severity scores to psychiatric symptoms based on intake reports.
• To create a concise, concept-based representation of patient information from clinical narratives.
• To assess the performance of this method in the context of the CEGS N-GRID 2016 Shared Task.

Main Methods:

• Utilized the interview-like structure of psychiatric intake reports to extract relevant information.
• Generated a representation based on a restricted set of psychiatric concepts linked to UMLS and DSM-IV.
• Employed Random Forests for generalization of case-specific features.
• Incorporated identification of concept certainty and scope (patient, family).

Main Results:

• Achieved an inverse mean absolute error (MAE) of 80.64% with the best performing model variant.
• Demonstrated that a concise, concept-based representation is effective for this task.
• Showcased the robustness of the method in identifying concept certainty and scope.

Conclusions:

• The proposed method effectively extracts and represents information from psychiatric intake reports for symptom severity scoring.
• Concept-based representations, including certainty and scope, offer a robust approach to clinical NLP tasks.
• This work contributes to advancing automated analysis of psychiatric clinical notes.